An intelligent diagnosis system and diagnosis process for hydraulic machine faults based on adaptive sound features

Abstract

This invention relates to the field of hydraulic press technology, and more particularly to an intelligent fault diagnosis system and process for hydraulic presses based on adaptive sound characteristics. The system includes a hydraulic press and a host computer control panel electrically connected to the hydraulic press. The host computer control panel is located on the side of the hydraulic press and includes a hydraulic control system and a diagnostic system. Different colored indicator lights are mounted on the top of the host computer control panel, and these indicator lights are connected to the host computer control panel. The diagnostic system is used to identify different sounds from the hydraulic press and determine whether a fault has occurred. This invention achieves the goal of being able to determine whether a hydraulic press is faulty by detecting subtle sounds that differ from the normal operating sound of the hydraulic press.

Description

Technical Field

[0001] This invention relates to the field of hydraulic press technology, and in particular to an intelligent fault diagnosis system and diagnostic process for hydraulic presses based on adaptive sound characteristics. Background Technology

[0002] Currently, hydraulic presses are crucial equipment in industrial production, and their operating status directly impacts production efficiency and product quality. Hydraulic press malfunctions can range from minor production line shutdowns to serious safety accidents. Therefore, accurately assessing the operating status of hydraulic presses and enabling early fault diagnosis and prevention are essential for improving production efficiency and ensuring safety. Experienced workers can often judge the operating status of a hydraulic press by the sound it makes; however, this method is highly dependent on personal experience, subjective, and lacks real-time, continuous monitoring capabilities, offering limited early warning for sudden or potential malfunctions.

[0003] In recent years, some fault identification technologies based on sound signals have emerged. These systems typically collect a batch of sound samples under normal and abnormal conditions in advance to establish a fixed diagnostic model. In practical applications, the real-time collected sound is compared with the model to determine the fault. However, this method has a significant drawback: a lack of adaptability. During the service life of a hydraulic press, its acoustic characteristics will inevitably evolve slowly due to normal wear and tear, changes in lubrication conditions, replacement of parts, or environmental interference. A fixed acoustic model established in the initial stage cannot track this dynamic change and will gradually become "unsuitable" for the current equipment state, resulting in a decrease in diagnostic accuracy and an increase in false alarms and missed alarms over time.

[0004] Therefore, this invention proposes an intelligent fault diagnosis system and process for hydraulic presses based on adaptive sound features, which can determine whether a hydraulic press is faulty by detecting subtle sounds that are different from the normal operating sound of the hydraulic press. Summary of the Invention

[0005] The purpose of this invention is to solve the problems existing in the prior art by proposing an intelligent fault diagnosis system and diagnosis process for hydraulic presses based on adaptive sound features.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent fault diagnosis and prediction system for hydraulic presses based on adaptive voiceprint recognition, comprising a hydraulic press and a host computer control panel and a hydraulic control system connected to the hydraulic press, wherein the host computer control panel integrates: A diagnostic system is used to monitor and assess the health status of the hydraulic press; the diagnostic system is configured with: A sound acquisition unit is used to acquire audio signals during the operation of the hydraulic press; An adaptive diagnostic unit, including an online-updable reference voiceprint model, processes and identifies the audio signal to determine the health status and fault type of the hydraulic press. A lifespan prediction unit is used to predict the remaining lifespan of key components based on the historical evolution data of the benchmark acoustic signature model. The feedback unit is used to output an alarm signal based on the diagnostic results of the adaptive diagnostic unit and the prediction results of the life prediction unit.

[0007] Furthermore, the adaptive diagnostic unit includes a sound processing module, a sound learning module, and a sound recognition module connected in sequence; The sound processing module is configured to: perform preprocessing and feature extraction on the audio signal acquired by the sound acquisition unit, and output a standardized sound feature vector; wherein, the preprocessing includes at least filtering and noise reduction, and the feature extraction includes at least calculating Mel-frequency cepstral coefficients; The sound learning module includes an online adaptive learning module; The voice recognition module is configured to: call the reference voiceprint model, calculate the matching degree between the real-time voice feature vector and the reference voiceprint model, and determine the health status and fault type of the hydraulic press based on whether the matching degree exceeds a preset threshold.

[0008] Furthermore, the online adaptive learning module is configured as follows: During the diagnostic system initialization phase, an initial reference acoustic signature model is established based on the original audio signal from the normal operation of the hydraulic press. During the online operation of the diagnostic system, new sound data under normal operating conditions that are determined by the sound recognition module to have a high degree of confidence are continuously collected; The baseline voiceprint model is dynamically optimized and updated using the new voice data through incremental learning or periodic retraining.

[0009] Furthermore, the high confidence level means that the confidence score representing the normal state in the diagnostic results output by the voice recognition module is higher than a preset safety threshold, which is set between 0.80 and 0.95.

[0010] Furthermore, the output of the feedback unit is connected to an indicator light and a display screen on the host computer operating console, respectively. The display screen is used to display fault information and remaining lifespan.

[0011] Furthermore, the lifetime prediction unit executes a performance degradation trend extrapolation algorithm, the core prediction formula of which is: ,in, RUL stands for predicted remaining useful life; T current This represents the current cumulative operating time of this critical component. P current These are the current values ​​of the key parameters characterizing the performance degradation of this component, extracted from the current benchmark acoustic signature model. P initial This is the initial value for this key parameter; T Threshold The failure threshold is preset based on experience; k is an acceleration factor related to the component material and wear characteristics.

[0012] Furthermore, a diagnostic process, applied to the aforementioned intelligent fault diagnosis system for hydraulic presses based on adaptive voiceprint recognition, includes the following steps: S1. Audio signal acquisition and preprocessing: Audio signals generated during the operation of the hydraulic press are acquired by a sound acquisition unit located near the hydraulic press; the audio signals are transmitted to the sound processing module of the diagnostic system for noise reduction and filtering preprocessing to obtain the audio signal to be analyzed. S2. Audio signal feature extraction: The audio signal to be analyzed is sent to the sound processing module of the diagnostic system for feature extraction; the sound processing module extracts the time domain, frequency domain and time-frequency features of the voiceprint signal in parallel, constructs an initial feature set, and generates a real-time sound feature vector representing the current state of the hydraulic press. S3. Adaptive Diagnostic Learning and Judgment: The real-time sound feature vector obtained in step S2 is used for diagnostic analysis within the diagnostic system; the sound recognition module in the diagnostic system calls the built-in benchmark voiceprint model for comparison and recognition, and outputs the preliminary diagnostic results and their corresponding confidence levels; if the confidence level is higher than the preset safety threshold and is determined to be in a normal state, the sound learning module is prepared to be activated; if it is determined to be in an abnormal state, the alarm output in step S5 is executed. S4. Adaptive learning of the baseline voiceprint model: When the learning conditions are met, the voice learning module is activated, retrieves the high-confidence voice feature vector of the current time, and optimizes the baseline voiceprint model through incremental learning or retraining to generate an updated baseline voiceprint model and replace the old baseline voiceprint model. S5. Diagnostic Result Output and Life Prediction: The final diagnostic result is displayed on the screen of the host computer console and alarms are triggered by the feedback unit and indicator lights. At the same time, the diagnostic information is sent to the life prediction unit to predict the remaining life of the key components of the hydraulic press. The high-confidence normal sound feature vector in this diagnosis is fed back and stored for the continuous learning and evolution of the diagnostic system.

[0013] Compared with existing technologies, the present invention provides an intelligent fault diagnosis system and diagnostic process for hydraulic presses based on adaptive sound features, which has the following advantages: 1. Improved diagnostic accuracy: By utilizing an online adaptive learning module to analyze the noise generated during the operation of the hydraulic press, the presence and type of faults in the hydraulic press can be identified more accurately. This algorithm can learn from large amounts of data and identify complex patterns, thereby improving the accuracy of fault diagnosis; 2. Real-time monitoring and early warning: The sound recognition module can process the noise data of the hydraulic press in real time, enabling real-time monitoring of the hydraulic press's status. Once abnormal noise is detected, the system can immediately issue an early warning to avoid potential equipment damage that could lead to production interruptions. 3. Reduce downtime: Early diagnosis and warning can reduce downtime of hydraulic presses, quickly locate problems and take measures, thereby reducing maintenance costs and improving production efficiency; 4. Reduced maintenance costs: By simply identifying the sound to inspect the operation of the hydraulic press, faults can be identified, reducing reliance on professional technicians and lowering maintenance and repair costs. 5. Improve safety: Timely identification and response to hydraulic press malfunctions can prevent accidents and improve workplace safety. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the present invention; Figure 2 This is a schematic diagram of the system framework structure of the present invention; Figure 3 This is a schematic diagram of the diagnostic process of the present invention; Figure 4 This is a flowchart of the diagnostic system of the present invention; Figure 5 This is a schematic diagram of the adaptive diagnostic unit framework of the present invention; In the diagram: 1. Host computer control panel; 11. Display screen; 2. Hydraulic control system; 3. Diagnostic system; 4. Sound acquisition unit; 5. Adaptive diagnostic unit; 51. Sound processing module; 52. Sound learning module; 520. Online adaptive learning module; 53. Sound recognition module; 6. Lifespan prediction unit; 7. Feedback unit; 8. Indicator light; 9. Hydraulic press; 10. Electrical control cabinet. Detailed Implementation

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0017] Example 1, as Figures 1-2 As shown, a hydraulic press fault intelligent diagnosis system based on adaptive sound features includes a hydraulic press 9, a host computer control panel 1 connected to the hydraulic press 9, and a hydraulic control system 2. The host computer control panel 1 integrates the following: Diagnostic system 3 is used to monitor and assess the health status of hydraulic press 9; Diagnostic system 3 is configured with: The sound acquisition unit 4 is used to acquire the audio signal when the hydraulic press 9 is running; The adaptive diagnostic unit 5 includes an online-updable reference voiceprint model that processes and identifies audio signals to determine the health status and fault type of the hydraulic press 9. The life prediction unit 6 is used to predict the remaining service life of key components based on historical evolution data of the benchmark acoustic model. Feedback unit 7 is used to output an alarm signal based on the diagnostic results of adaptive diagnostic unit 5 and the prediction results of life prediction unit 6.

[0018] In this embodiment, the system includes a hydraulic press 9 as the monitoring object and a host computer console 1 physically located adjacent to it. An electrical control cabinet 10, actuators, and a hydraulic control system 2 are connected to the hydraulic press 9. The operation of the hydraulic press 9 is controlled by the electrical control cabinet 10 and the hydraulic control system 2. A diagnostic system 3 forms a closed-loop control link with the electrical control cabinet 10 and the hydraulic control system 2 through a feedback unit 7. The electrical control cabinet 10 has a built-in PLC controller. The actuators include hydraulic pumps, hydraulic cylinders, etc. The hydraulic control system 2 includes control valve groups (such as directional valves, pressure valves, flow valves), oil tanks, pipelines, seals, etc. The host computer console 1 establishes an electrical connection with the PLC controller of the hydraulic press 9 through an internal data bus or communication interface (such as Ethernet, PROFIBUS, etc.) to obtain the basic operating status of the hydraulic press (such as start-up, stop, working pressure, etc.).

[0019] The sound acquisition unit 4 includes a high-precision sound and vibration acquisition unit that integrates acoustics and vibration. This high-precision sound and vibration acquisition unit can acquire the audio signals of the hydraulic press in real time and provide timely feedback on the equipment status. The sensor in the high-precision sound and vibration acquisition unit has high sensitivity and can detect minute changes in sound and vibration. It can be installed at the locations on the hydraulic press 9 where monitoring is required, as needed.

[0020] It can be installed in the cylinders and sliders of the hydraulic press 9; the oil pumps and oil pipes of the hydraulic control system 2; electrical components of the electrical control system, such as relays and contactors; and other auxiliary components, such as cooling systems and lubrication systems.

[0021] After acquiring audio signals from various locations on the hydraulic press 9, the high-precision sound vibration acquisition device continuously transmits the audio signals to the sound processing module 51 via a network interface.

[0022] The host computer control console 1 is a central hardware unit integrating computing, storage, control, and display functions. Its hardware components and collaborative operation are as follows: Industrial computer: As the core computing hub, it runs an operating system (such as Linux) and application software such as diagnostic system 3 and hydraulic control system 2 running on the operating system.

[0023] Data acquisition card: As a dedicated signal acquisition hardware, its input end is directly and physically connected to all high-precision sound and vibration acquisition devices on the hydraulic press 9 through a shielded signal cable, and is responsible for converting continuous time-varying voltage signals into digital signals; its output end is connected to the industrial computer through a PCIe interface, and transmits the digital signals to the memory, which are then called and processed by modules such as the sound processing module 51 in the diagnostic system 3.

[0024] The host computer control panel 1 also has a display screen 11, which serves as the human-machine interface and is connected to the graphics card interface of the industrial computer via a cable. It can also input parameters of the hydraulic control system 2 during operation, such as pressure setpoints and flow control parameters. Furthermore, it can display real-time fault prompts, facilitating quick problem location and resolution by operators.

[0025] In terms of hardware connection, the sound acquisition unit 4 (i.e., the high-precision sound vibration acquisition unit) is directly and physically connected to the data acquisition card via a shielded signal cable, and outputs a continuous time-varying voltage signal. The data acquisition card converts this time-varying voltage signal into a digital signal.

[0026] In terms of software logic, the sound processing module 51 in the diagnostic system 3 receives the digital signal by calling the API of the data acquisition card and uses it as the input for subsequent feature extraction and diagnostic analysis.

[0027] The adaptive diagnostic unit 5 is a software algorithm module whose input terminal receives digital signals from the sound acquisition unit 4 through a software interface.

[0028] The baseline acoustic model is stored in one or more data files in the non-volatile memory (such as hard disk or flash memory) of the industrial computer inside the host computer console 1. The data file records a set of mathematical parameters learned by the acoustic learning module 52 that can uniquely characterize the acoustic characteristics of the hydraulic press 9 in a healthy state.

[0029] Feedback unit 7 is also a software module. Its input end receives diagnostic results (such as "normal", "pump failure", "valve failure") from adaptive diagnostic unit 5 through inter-process communication. Its output end sends switching commands (such as "light up the red indicator") to the drive circuit of indicator light 8 through system calls and graphical interface, and sends data to the graphical interface of display screen 11 to pop up alarm information.

[0030] Specifically, the operating system calls and the graphical interface are used to implement low-level hardware control of indicator light 8. The software indirectly operates its drive circuit by issuing instructions to the operating system.

[0031] The graphical interface is used to enable human-computer interaction with the display screen 11. The feedback unit 7 renders diagnostic data and alarm information on the screen by calling the API provided by the graphical user interface framework. The API (Application Programming Interface) is a set of predefined calling specifications and function routines provided by the operating system for upper-level application software. By calling these APIs, the software modules in the diagnostic system 3 (such as the feedback unit 7) can request and obtain specific services without understanding the complex underlying implementation details.

[0032] With this technical solution, the hydraulic press 9 will produce different frequencies of sound when different faults occur. The host computer control panel 1 receives the different sounds produced by the hydraulic press 9 and can then determine where the fault is. When different faults occur, the indicator light 8 will also play an alarm role.

[0033] Example 2, continue to refer to Figure 2 As shown, the adaptive diagnostic unit 5 includes a sound processing module 51, a sound learning module 52, and a sound recognition module 53 connected in sequence. The sound processing module 51 is configured to: perform preprocessing and feature extraction on the audio signal acquired by the sound acquisition unit 4, and output a standardized sound feature vector; wherein, the preprocessing includes at least filtering and noise reduction, and the feature extraction includes at least calculating the Mel frequency cepstral coefficients; The sound learning module 52 includes an online adaptive learning module 520; The sound recognition module 53 is configured to: call the reference voiceprint model, calculate the matching degree between the real-time sound feature vector and the reference voiceprint model, and determine the health status and fault type of the hydraulic press 9 based on whether the matching degree exceeds a preset threshold.

[0034] In this embodiment, refer to Figure 4 As shown, the diagnostic system 3 mainly includes two stages: diagnostic system 3 initialization and diagnostic system 3 online operation and adaptation. In the first stage, the diagnostic system 3 is initialized, such as when it is first used, after a major overhaul of the hydraulic press or after the replacement of major components. The goal is to obtain a clean "healthy state" baseline voiceprint model.

[0035] First, data acquisition is performed to ensure that the hydraulic press 9 is in good working order. Data is then collected under multiple typical working cycles of the hydraulic press 9. These conditions should cover the normal operating pressure range and speed range of the hydraulic press 9.

[0036] Next, data preprocessing and feature extraction are performed. The acquired raw audio signal is transmitted to the sound processing module 51. The sound processing module 51 is an integrated signal processor that can contain multiple processing steps or logic submodules, including a signal preprocessing submodule, a feature extraction submodule, and an output submodule. The signal preprocessing submodule can use existing technologies to filter and denoise the raw audio signal, such as deep learning-based denoising algorithms, to filter out stable background noise, such as fan noise and other equipment noise. A bandpass filter is used to retain the main frequency band of the hydraulic press's operating sound while removing irrelevant high-frequency and low-frequency interference. Then, the continuous audio signal is segmented into short time frames for subsequent time-frequency analysis, resulting in a set of clean, fixed-length short-time audio signal frames, i.e., the audio signal to be analyzed.

[0037] The feature extraction submodule extracts features from short-time audio signal frames. It can use methods such as calculating Mel-frequency cepstral coefficients, root mean square values, and fast Fourier transform to extract features, obtaining a trainable sound feature vector for the hydraulic press 9 under normal conditions.

[0038] All extracted trainable sound feature vectors are input to the sound learning module 52 through the output submodule. The online adaptive learning module 520 of the sound learning module 52 receives the sound feature vector sequence. It calls the Gaussian mixture model training algorithm and iteratively calculates using the expectation-maximization algorithm to finally generate an initial benchmark voiceprint model composed of K Gaussian distributed component parameters (weights, mean vector, covariance matrix). The initial benchmark voiceprint model is persistently stored in the memory of the host computer console 1. In another embodiment, the initial benchmark voiceprint model can also be a deep autoencoder, trained by minimizing the reconstruction error through the backpropagation algorithm, and its final network weights are the initial benchmark voiceprint model.

[0039] After training, an initial baseline voiceprint model is generated. Essentially, this model defines the probability distribution range of the sound feature vector representing the normal state of the hydraulic press 9 in a multi-dimensional space. Any new audio signal, after undergoing the same feature extraction process, can have its "degree of conformity" or "degree of deviation" from this baseline voiceprint model calculated.

[0040] Similarly, during the online operation of the diagnostic system 3, it enters the online monitoring mode, and the sound acquisition unit 4 continuously acquires the operating sound of the hydraulic press 9, i.e., the real-time audio signal. The sound processing module 51 performs noise reduction and filtering on the real-time audio signal using the same parameters as described above. A real-time sound feature vector is formed and output to the sound recognition module 53.

[0041] The voice recognition module 53 matches the real-time voice feature vector with the current initial baseline voiceprint model. The voice recognition module 53 calculates a confidence score or anomaly score. This score indicates the probability that the current voice belongs to the "normal" range.

[0042] The decision logic is as follows: If the confidence score is higher than the preset safety threshold, it is judged as "normal operation"; If the confidence score is lower than the safety threshold, it is judged as "suspected abnormality" and an alarm is issued through the display screen 11 and indicator light 8 of the host computer console 1.

[0043] High confidence means that the confidence score of the diagnostic results output by the voice recognition module 53, which represents a normal state, is higher than a preset safety threshold, which is set between 0.80 and 0.95.

[0044] For example, the preset safety threshold is determined based on the confidence distribution calculated from the original sound signals collected during the initialization phase of the diagnostic system 3 under normal operating conditions of the hydraulic press 9; the safety threshold ranges from 0.80 to 0.95. Preferably, the safety threshold is set to 0.85-0.90. If the confidence level is higher than this safety threshold, it is determined to be "normal". If the confidence level is lower than this safety threshold, it is determined to be "abnormal" and an alarm is triggered. Those skilled in the art can manually set and fine-tune the safety threshold within the recommended range through the display screen 11 of the host computer control panel 1, according to the criticality of the hydraulic press 9 and the strictness of the process requirements.

[0045] like Figure 5As shown, the diagnostic system 3 also includes an adaptive learning buffer in its software architecture. This adaptive learning buffer is physically located in the memory of the host computer console 1 and is a first-in, first-out queue created and managed by the voice learning module 52. Its core function is to temporarily store high-confidence normal voice feature vectors provided by the voice recognition module 53, whose confidence level is higher than the safety threshold, thereby providing a high-quality data source for the safe and batch updates of the baseline voiceprint model.

[0046] When the hydraulic press 9 is determined to be in normal condition, the following adaptive process is executed: a. Store these new normal sound feature vector data into an adaptive learning buffer; b. The sound learning module 52 reads data from the adaptive learning buffer and executes the model update algorithm; c. Use a model update algorithm to dynamically optimize and update the baseline voiceprint model stored in the memory of the diagnostic system 3.

[0047] When the hydraulic press 9 is determined to be in an abnormal state, the abnormal signal of the hydraulic press is transmitted to the feedback unit 7.

[0048] The output of the feedback unit 7 is connected to an indicator light 8 installed on the host computer console 1; the output of the feedback unit 7 is also connected to the display screen 11 of the host computer console 1, which is used to display fault information and remaining lifespan.

[0049] Indicator light 8 is installed on the top panel of the control panel 1 and is connected to the output interface of the industrial computer or the digital output module of the PLC via internal wires to receive alarm signals.

[0050] For example, when air enters the hydraulic cylinder 9, it will generate noise or heat. If the hydraulic press 9 is running at this time, the sound from the cylinder will be different from the normal operating sound. After the high-precision sound and vibration collector collects the operating sound from the cylinder, it transmits this sound to the sound processing module 51. As another example, when an oil pipe leaks, it will emit a specific sound. The high-precision sound and vibration collector transmits the leak sound to the sound processing module 51, allowing for early intervention to address the oil pipe leak problem.

[0051] The sound processing module 51 receives and processes audio signals from multiple high-precision sound vibration acquisition units (sound acquisition units 4) at various locations on the hydraulic press 9. The sound processing module 51 first preprocesses each audio signal, using filters available in the prior art to remove background noise, such as spectral subtraction and Wiener filtering. Then, it extracts features and fuses them into a comprehensive real-time sound feature vector. The sound recognition module 53 then calls a benchmark voiceprint model, compares the comprehensive real-time sound feature vector with the benchmark voiceprint model, and outputs the diagnostic result.

[0052] A normal range of values ​​can be defined for these audio frequencies. If the sound recognition module 53 identifies that the collected sound is greater than or less than this normal range, it will be considered an abnormal sound. In this case, feedback will be given through the feedback unit 7, which will send the fault information of the hydraulic press 9 to the host computer control panel 1. The display screen 11 on the host computer control panel 1 will then display the relevant fault information prompts of the hydraulic press 9. At the same time, the feedback unit 7 will also transmit the relevant fault information of the hydraulic press 9 to the indicator lights 8. The indicator lights 8 will issue alarm prompts according to the different faults of the hydraulic press 9. In some embodiments, we can sort the indicator lights. For example, if the indicator light 8 with the serial number 1 is lit, it indicates that the oil cylinder is faulty. If the indicator light 8 with the serial number 2 is lit, it indicates that the oil pump is abnormal.

[0053] Indicator 8 can be set to either a lighting mode or a buzzer mode. In either mode, the light and sound can alert the operator when the operator is at a distance from the hydraulic press 9.

[0054] Indicator light 8 and display screen 11 can simultaneously display fault messages for hydraulic press 9, ensuring the accuracy and real-time transmission of fault information for hydraulic press 9. This is to prevent situations where one of indicator light 8 or display screen 11 malfunctions and the staff fails to detect it in time.

[0055] In Example 3, to predict the service life of various components, a service life prediction unit 6 is included in the diagnostic system 3. The service life prediction unit 6 executes a performance degradation trend extrapolation algorithm. The core of this algorithm lies in using historical data from a benchmark acoustic signature model to predict the remaining service life of key components through a nonlinear empirical model. To achieve accurate service life prediction, this invention defines and tracks a core indicator called "Key Parameter P". This "Key Parameter P" quantitatively characterizes the performance degradation state of a specific component of the hydraulic press (such as the main hydraulic pump). "Key Parameter P" is a scalar value derived from the benchmark acoustic signature model. Its physical essence is the acoustic energy amplitude within a specific characteristic frequency band that is strongly correlated with the mechanical wear of the target component. The value of "Key Parameter P" monotonically increases with the increase of wear on the target component, thus providing a stable and reliable observation window for quantifying performance degradation.

[0056] Based on the "critical parameter P", we defined three decisive state points for it throughout the entire component lifecycle: the initial value of the "critical parameter P" P initial The current value of the "key parameter P" P current Failure threshold T of "critical parameter P" Threshold Its core prediction formula is: ,in, RUL stands for predicted remaining useful life; T currentThis is the current cumulative operating time of the critical component; it can be directly read from the hydraulic control system 2 of the host computer console 1 or the PLC in the electrical control cabinet 10. P current These are the current values ​​of the key parameters characterizing the performance degradation of this component, extracted from the current benchmark acoustic signature model. Specifically, when extracting real-time audio signals, the sound processing module 51 calculates the energy amplitude (e.g., the root mean square value of the 500Hz to 2000Hz band) of a characteristic frequency band strongly correlated with the wear of the target component (such as a hydraulic pump). This calculated amplitude is recorded and updated in the current reference acoustic signature model as an explicit, accessible parameter. The lifespan prediction unit 6 can read the latest P by accessing the current reference acoustic signature model. current This value reflects the intensity of abnormal vibrations and noise generated by wear on the component at the current moment. The more severe the wear, the larger the value. P initial This is the initial value for this key parameter; during the initialization of the diagnostic system 3, the hydraulic press 9 is in a brand-new or overhauled good condition. The energy amplitude of the specific characteristic frequency band most relevant to the wear of the target component, extracted from the initial baseline acoustic signature model established at this time, is P. initial This is a fixed reference value, permanently stored in the internal memory of the host computer console 1; T Threshold The failure threshold is preset based on experience; T hreshold These are critical values ​​for key parameters, preset based on experience, used to determine whether a component has failed. According to industry standards or service manuals, this is the threshold value when the characteristic energy amplitude reaches a certain level (e.g., compared to its initial value P). initial When the voltage is 20 dB higher than normal, the component is considered to have failed; the corresponding value for this level is T. hreshold ; k is an acceleration factor related to the component material and wear characteristics.

[0057] It can be determined by curve fitting of historical full life cycle data of the same model of equipment.

[0058] The workflow for lifetime prediction: Step 1: Parameter Preparation and Acquisition Read performance degradation parameters: The lifetime prediction unit 6 retrieves the current baseline voiceprint model maintained by the voice learning module 52 from the system memory; Extract the current value P of the predefined key parameter characterizing component performance degradation from the current benchmark acoustic signature model. current ; Calling benchmark and empirical parameters: Read the initial value P of the key parameter of this component from the internal memory. initial This value originates from the initial baseline voiceprint model established during the initialization phase; At the same time, read the preset failure threshold T hreshold and acceleration factor k; Get runtime: The current cumulative running time T of this critical component can be read from the PLC in the hydraulic control system 2 or the electrical control cabinet 10 via a software interface. current ; Execute the core algorithm: Lifetime prediction unit 6 substitutes all the parameters obtained in the above steps into the core formula of the performance degradation trend extrapolation algorithm: The calculation is completed to obtain the predicted remaining useful life (RUL). Step Two: Result Feedback and Early Warning The lifetime prediction unit 6 sends the calculated predicted remaining useful life (RUL) value to the feedback unit 7. Step 3: Human-computer interaction and early warning: Feedback unit 7 drives the display screen 11 of the host computer console 1 to display the remaining lifespan in a graphical manner, such as progress bars, numerical values, trend curves, etc.

[0059] When the remaining service life (RUL) value is lower than the preset warning value, the feedback unit 7 simultaneously drives the indicator light 8 to emit a specific color light (such as yellow) to issue a warning, and converts it into an alarm signal (such as red) when the remaining service life (RUL) value is close to zero.

[0060] Example 4, a diagnostic procedure, such as Figures 3-5 As shown, the hydraulic press fault diagnosis system based on adaptive sound features described above includes the following steps: S1. Audio signal acquisition and preprocessing: The audio signal generated by the hydraulic press 9 during operation is acquired by the sound acquisition unit 4 arranged near the hydraulic press 9; the original audio signal is transmitted to the sound processing module 51 in the diagnostic system 3 for noise reduction and filtering preprocessing to obtain the audio signal to be analyzed. S2. Audio signal feature extraction: The audio signal to be analyzed is sent to the sound processing module 51 of the diagnostic system 3 for feature extraction; the sound processing module 51 extracts the time domain, frequency domain and time-frequency features of the voiceprint signal in parallel, constructs an initial feature set, and generates a real-time sound feature vector characterizing the current state of the hydraulic press 9. S3. Adaptive Diagnostic Learning and Judgment: The real-time sound feature vector obtained in step S2 is analyzed in the diagnostic system 3. The sound recognition module 53 in the diagnostic system 3 calls the built-in benchmark voiceprint model for comparison and recognition, and outputs the preliminary diagnostic results and their corresponding confidence levels. If the confidence level is higher than the preset safety threshold and is determined to be in a normal state, the sound learning module 52 is prepared to be activated. If it is determined to be in an abnormal state, the alarm output in step S5 is executed. When the voice recognition module 53 determines that the current state is normal and the confidence level is higher than the safety threshold, the diagnostic system 3 will not immediately update the baseline voiceprint model, but will instead store the real-time voice feature vector at that moment, along with its timestamp and confidence level metadata, into a dedicated adaptive learning buffer.

[0061] The adaptive learning buffer is a first-in, first-out queue allocated in the memory of the diagnostic system 3, configured to store 500-1000 sound feature vectors. As a data bridge connecting the diagnostic and learning functions, the adaptive learning buffer has the following management mechanisms: 1. Entry Mechanism: Only sound feature vectors with a confidence level higher than a specific threshold can enter the buffer. Within the range of diagnostic alarm thresholds, a higher value (e.g., 0.90) is selected as the data admission threshold to control the quality of data entering the adaptive learning buffer. Only when a real-time sound feature vector is determined to be "normal" and its confidence level is higher than 0.90 will the data be stored in the adaptive learning buffer for the sound learning module 52 to use to update the baseline voiceprint model. 2. Overflow Management: When the adaptive learning buffer reaches its capacity limit, the data that has been stored for the longest time is automatically removed; 3. Quality maintenance: The diagnostic system periodically performs consistency checks on the data in the adaptive learning buffer and removes sound feature vectors that deviate significantly from the current data distribution.

[0062] S4. Adaptive learning of the baseline voiceprint model: When the learning conditions are met, the learning conditions are the management mechanism of the adaptive learning buffer in S3. The sound learning module 52 is activated, and the high-confidence sound feature vector of this time is retrieved. The baseline voiceprint model is optimized through incremental learning or retraining to generate an updated baseline voiceprint model and replace the old baseline voiceprint model. S5. Diagnostic Result Output and Life Prediction: The final diagnostic result is displayed on the screen 11 of the host computer console 1 and alarms are triggered by the feedback unit 7 and indicator lights 8. At the same time, the diagnostic information is sent to the life prediction unit 6 to predict the remaining life of the key components of the hydraulic press 9. The high-confidence normal sound feature vector in this diagnosis is fed back and stored for the continuous learning and evolution of the diagnostic system 3.

[0063] In this embodiment, the host computer control console 1 can also be remotely controlled via a mobile terminal, which can be a mobile phone, tablet, etc.

[0064] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A hydraulic press fault intelligent diagnosis system based on adaptive sound features, comprising a hydraulic press (9) and a host computer operating console (1) and a hydraulic control system (2) connected to the hydraulic press (9), characterized in that: The host computer control panel (1) integrates the following: A diagnostic system (3) is used to monitor and evaluate the health status of the hydraulic press (9); the diagnostic system (3) is configured with: The sound acquisition unit (4) is used to acquire the audio signal of the hydraulic press (9) during operation; The adaptive diagnostic unit (5) includes an online updatable reference voiceprint model, which processes and identifies the audio signal to determine the health status and fault type of the hydraulic press (9). The life prediction unit (6) is used to predict the remaining service life of key components based on the historical evolution data of the benchmark acoustic model. Feedback unit (7) is used to output an alarm signal based on the diagnostic results of adaptive diagnostic unit (5) and the prediction results of life prediction unit (6).

2. The intelligent fault diagnosis system for hydraulic presses based on adaptive sound features according to claim 1, characterized in that: The adaptive diagnostic unit (5) includes a sound processing module (51), a sound learning module (52), and a sound recognition module (53) connected in sequence. The sound processing module (51) is configured to: perform preprocessing and feature extraction on the audio signal acquired by the sound acquisition unit (4) and output a standardized sound feature vector; wherein the preprocessing includes at least filtering and noise reduction, and the feature extraction includes at least calculating the Mel frequency cepstral coefficients; The sound learning module (52) includes an online adaptive learning module (520); The sound recognition module (53) is configured to: call the reference voiceprint model, calculate the matching degree between the real-time sound feature vector and the reference voiceprint model, and determine the health status and fault type of the hydraulic press (9) based on whether the matching degree exceeds a preset threshold.

3. The intelligent fault diagnosis system for hydraulic presses based on adaptive sound features according to claim 2, characterized in that: The online adaptive learning module (520) is configured as follows: During the initialization phase of the diagnostic system (3), an initial reference acoustic model is established based on the original audio signal of the hydraulic press (8) during normal operation; During the online operation of the diagnostic system (3), new sound data under normal operating conditions that are determined by the sound recognition module (53) to have high confidence are continuously collected; The baseline voiceprint model is dynamically optimized and updated using the new voice data through incremental learning or periodic retraining.

4. The intelligent fault diagnosis system for hydraulic presses based on adaptive sound features according to claim 3, characterized in that: The high confidence level refers to the fact that the confidence score representing the normal state in the diagnostic results output by the sound recognition module (53) is higher than a preset safety threshold, which is set between 0.80 and 0.

95.

5. The intelligent fault diagnosis system for hydraulic presses based on adaptive sound features according to claim 1, characterized in that: The output of the feedback unit (7) is connected to the indicator light (8) and the display screen (11) set on the host computer operating console (1), respectively. The display screen (11) is used to display fault information and remaining life.

6. The intelligent fault diagnosis system for hydraulic presses based on adaptive sound features according to claim 1, characterized in that: The lifetime prediction unit (6) executes a performance degradation trend extrapolation algorithm, and its core prediction formula is: ,in, RUL stands for predicted remaining useful life; T current This represents the current cumulative operating time of this critical component. P current These are the current values ​​of the key parameters characterizing the performance degradation of this component, extracted from the current benchmark acoustic signature model. P initial This is the initial value for this key parameter; T Threshold The failure threshold is preset based on experience; k is an acceleration factor related to the component material and wear characteristics.

7. A diagnostic process applied to the intelligent fault diagnosis system for hydraulic presses based on adaptive sound features as described in claims 1-6, characterized in that: Includes the following steps: S1. Audio signal acquisition and preprocessing: The audio signal generated by the hydraulic press (9) during operation is acquired by the sound acquisition unit (4) arranged near the hydraulic press (9); the audio signal is transmitted to the sound processing module (51) in the diagnostic system (3) for noise reduction and filtering preprocessing to obtain the audio signal to be analyzed. S2, Audio signal feature extraction: The audio signal to be analyzed is sent to the sound processing module (51) of the diagnostic system (3) for feature extraction; the sound processing module (51) extracts the time domain, frequency domain and time-frequency features of the voiceprint signal in parallel, constructs an initial feature set, and generates a real-time sound feature vector representing the current state of the hydraulic press (9); S3. Adaptive diagnostic learning and judgment: The real-time sound feature vector obtained in step S2 is used for diagnostic analysis in the diagnostic system (3); the sound recognition module (53) in the diagnostic system (3) calls the built-in benchmark voiceprint model for comparison and recognition, and outputs the preliminary diagnostic results and their corresponding confidence levels; if the confidence level is higher than the preset safety threshold and is determined to be in a normal state, then the sound learning module (52) is prepared to be activated; if it is determined to be in an abnormal state, then the alarm output in step S5 is executed. S4, Adaptive learning of the baseline voiceprint model: When the learning conditions are met, the sound learning module (52) is activated, the high confidence sound feature vector of this time is retrieved, and the baseline voiceprint model is optimized by incremental learning or retraining, generating an updated baseline voiceprint model and replacing the old baseline voiceprint model. S5. Diagnostic results output and life prediction: The final diagnostic results are displayed on the screen (11) of the host computer console (1) and alarms are triggered by the feedback unit (7) and indicator lights (8); at the same time, the diagnostic information is sent to the life prediction unit (6) to predict the remaining life of the key components of the hydraulic press (9); the high confidence normal sound feature vector in this diagnosis is fed back and stored for the continuous learning and evolution of the diagnostic system (3).

Images